Unsupervised Learning on a DIET: Datum IndEx as Target Free of Self-Supervision, Reconstruction, Projector Head

Costly, noisy, and over-specialized, labels are to be set aside in favor of unsupervised learning if we hope to learn cheap, reliable, and transferable models. To that end, spectral embedding, self-supervised learning, or generative modeling have offered competitive solutions. Those methods however come with numerous challenges textite.g. estimating geodesic distances, specifying projector architectures and anti-collapse losses, or specifying decoder architectures and reconstruction losses. In contrast, we introduce a simple explainable alternative -- coined DIET -- to learn representations from unlabeled data, free of those challenges. DIET is blatantly simple: take one's favorite classification setup and use the Datum IndEx as its Target class, textiti.e. each sample is its own class, no further changes needed. DIET works without a decoder/projector network, is not based on positive pairs nor reconstruction, introduces no hyper-parameters, and works out-of-the-box across datasets and architectures. Despite DIET's simplicity, the learned representations are of high-quality and often on-par with the state-of-the-art textite.g. using a linear classifier on top of DIET's learned representation reaches 71.4% on CIFAR100 with a Resnet101, 52.5% on TinyImagenet with a Resnext50.